From 88e42178788bb0becf10e9ff1fc9e7a2f71db733 Mon Sep 17 00:00:00 2001
From: Saoud Rizwan <7799382+saoudrizwan@users.noreply.github.com>
Date: Fri, 30 Aug 2024 22:37:58 -0400
Subject: [PATCH] Set better max token size

---
 src/ClaudeDev.ts | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/src/ClaudeDev.ts b/src/ClaudeDev.ts
index 5fe2637..c1ea9be 100644
--- a/src/ClaudeDev.ts
+++ b/src/ClaudeDev.ts
@@ -1305,7 +1305,7 @@ ${this.customInstructions.trim()}
 `
 			}
 
-			// Check last API request metrics to see if we need to truncate
+			// If the last API request's total token usage is close to the context window, truncate the conversation history to free up space for the new request
 			const lastApiReqFinished = findLast(this.claudeMessages, (m) => m.say === "api_req_finished")
 			if (lastApiReqFinished && lastApiReqFinished.text) {
 				const {
@@ -1317,8 +1317,9 @@ ${this.customInstructions.trim()}
 					lastApiReqFinished.text
 				)
 				const totalTokens = (tokensIn || 0) + (tokensOut || 0) + (cacheWrites || 0) + (cacheReads || 0)
-				const isCloseToContextWindowLimit = totalTokens >= this.api.getModel().info.contextWindow * 0.8
-				if (isCloseToContextWindowLimit) {
+				const contextWindow = this.api.getModel().info.contextWindow
+				const maxAllowedSize = Math.max(contextWindow - 20_000, contextWindow * 0.8)
+				if (totalTokens >= maxAllowedSize) {
 					const truncatedMessages = truncateHalfConversation(this.apiConversationHistory)
 					await this.overwriteApiConversationHistory(truncatedMessages)
 				}